Scaling Up from Drone to Satellite for Monitoring Dryland Ecosystem Phenology

The landscape of drylands, the planet's largest terrestrial biome, is typically formed by discrete plant patches surrounded by communities of cyanobacteria, mosses and/or lichens (biocrusts). Monitoring the spatial distribution and the phenological cycle of these biotic components is important, as they play a major role in ecohydrological processes of water limited environment. We used a scaling-up approach integrating multi-source data (i.e., ground measurements, multispectral drone and ESA Sentinel-2 (S2) images) to monitor the temporal succession of typical dryland components (i.e., biocrusts and vascular plants) in a semi-arid ecosystem in Spain (Aranjuez, 40°01'53.9" N 3°32'50.8" W). The optical properties of 72 biocrust samples, collected in March 2018, were characterized in laboratory with a hyperspectral imaging spectrometer (400-1000 nm). The biocrust components were classified in each Petri through a support vector machine algorithm (overall accuracy = 98 %) and their fractional covers (Fc) were calculated. The drone images were used to map vascular plants, biocrusts and bare soils through a supervised approach. Furthermore, the linear correlation (R2 = 0.86) between moss Fc and the chlorophyll a absorption feature at 670 nm, calculated with the continuum removal technique (CR670), was used to define three classes of biocrusts: i) moss dominant, ii) lichen and moss mixed, and iii) lichen dominant. We then calculated CR670 from a dataset of S2 reflectance images (L2-A product) acquired in the rainy season. The S2 CR670 was analyzed in relation to the ground precipitation data from a nearby weather station and the ecosystem components. In particular, we focused on the change in CR670 values due to biocrust activation, occurring just after the first precipitation event, and vegetation development in relation to S2 pixel composition, i.e., Fc of moss, lichens, bare soil and vascular plants. Results showed that S2 CR670 can detect: 1) the biocrust activation when biocrust cover is dominated by mosses, while the detection is hard when lichens are dominant, 2) different phenological trends depending on the pixel composition. These results showed the potential of S2 for monitoring the impacts of climate change on the status of both vascular plants and biocrusts in dryland ecosystems.